Elsevier

Pattern Recognition

Volume 35, Issue 2, February 2002, Pages 465-472
Pattern Recognition

Non-contact velocity compensation system for handheld scanners

https://doi.org/10.1016/S0031-3203(01)00056-5Get rights and content

Abstract

In this paper, we present a method for compensating for variations in the scanning speed of hand-held scanners that does not require physical contact between the scanning device and the scanned surface. In existing scanning devices the scanning linear array is placed parallel to the paper width (main scan direction) and moves perpendicularly to its axis (sub-scan direction). The speed with which the array moves determines the scanning resolution along the sub-scan direction. If the scanner is moved manually as in a handheld scanner the velocity with which the array moves along the sub-scan direction varies, thus causing distortions of the scanned images. In this paper, we present a system that can estimate these velocity variations using local contrast estimates. This system is unique in that it does not require contact between image source and scanning head, it can therefore be used by visually impaired users in a variety of environments. The presented velocity compensation system provides a post-compensation (apparent) velocity in the range of 1.2 : 1 for an initial (real) velocity variation of 10 : 1.

Introduction

The introduction of small, handheld digital scanners allowed great flexibility to users that wanted to scan a variety of documents, that cannot be easily scanned with flat bed scanners. Handheld scanners, however, introduced new problems for optical character recognition researchers. All character recognition systems, independent of the recognition methods [1], [2], [3], [4], [5] that they employ, expect the aspect ratio of the scanned characters to remain relatively invariant. However, when a handheld scanner is used, the scanning velocity and direction are controlled by the user's hand motion; and they are not kept constant as in a mechanical page scanner. These velocity variations across one dimension of the scanned test can cause problems in optical character recognition systems.

One way to track the scanning velocity is to place some kind of ruler on the document and use the marks of the ruler to estimate the traveled distance. This approach was implemented in Ref. [7], but required precise and careful motion of the scanning wand; as a consequence, it was soon abandoned.

Another way of tracking the scanning velocity is to have a roller attached to the scanning array and measure the velocity as a function of the angular velocity of the roller [7], [8], [9], [10]. Such a solution requires a steady and constant contact between the roller and the scanned surface, conditions that cannot always be met. An open book, for example, does not provide a flat surface that can be easily scanned by such a device.

A third alternative is to use a two-dimensional array sensor instead of a linear sensor and take multiple “snapshots” of the text. When a character is completely inside the viewing area of the sensor, it is acquired and the process continues with the next character [6]. This approach presents several limitations; namely, the size of the scanned text is limited by the size of the array sensor. Since several snapshots must be processed for each character, the size of the two dimensional sensor must be small; otherwise, the recognition speed will be extremely slow, only one line of text can be processed at a time and graphics cannot be handled.

In this paper, we present a velocity compensation system which is based on a technique suggested by one of us [11] and which relies on the uniformity of print across a scanned document. This system does not require any physical contact between the scanning head and the document.

Section snippets

Velocity compensation principle

Most existing scanning devices use a linear photosensitive array which is placed parallel to the paper width (usually called the main scan direction) and moves perpendicularly to its axis (usually called the sub-scan direction) in such a way that the entire image is scanned when the array travels the entire length of the paper. The speed with which the array moves along the sub-scan direction is a function of the desired resolution and is kept constant so that the resolution along the two

Implemented algorithm

Let R[i,t] denote the ith pixel in the tth column of video. Then an estimate of the contrast along the y direction DY and an estimate of the contrast along the x direction DX is computed from the differences between adjacent pixels along the two directions as follows:

Velocity compensation algorithm (part A).

For every pixel in a line of video the following steps are executed:

Step 1. At the beginning of each line contrast estimations DX and DY are initialized to 0.

If i=0 or t=0 then DY=0 and DX

Hardware implementation

The above velocity compensation algorithm requires two types of computations: those that must be completed within one pixel time (all steps in part A) and computations that are required only between two successive lines of video (all steps in part B). Fig. 7 shows the detailed algorithmic sequence of part A as it was actually implemented, while Fig. 8 shows in detail the steps of part B of the algorithm as it was actually implemented.

It is obvious that the computations in part A require fast

Results and conclusions

This velocity compensation subsystem has been tested with several documents and illuminations. It has also been tested with a prototype scanning device that can scan documents with variable speeds along the sub-scan direction and with a hand-held scanner. The tests were very successful. For scanning velocities that vary between 10% and 100% of the nominal velocity, the post-compensation apparent velocities range between 90% and 100% of the nominal velocity. Tests with commercial OCR software

About the Author—NIKOS ASIMOPOULOS received the B.S. in Electrical Engineering from the Aristotelean University of Thessaloniki, Greece and the M.S. and Ph.D. degrees from Virginia Tech. In 1988, he joined Image Processing Technologies Inc. where he now holds the position of VP of Engineering. Since 1996 he has been teaching as adjunct professor at the Electrical and Computer Engineering Department of Virginia Tech. His interests include real time image processing and pattern recognition with

References (11)

There are more references available in the full text version of this article.

Cited by (0)

About the Author—NIKOS ASIMOPOULOS received the B.S. in Electrical Engineering from the Aristotelean University of Thessaloniki, Greece and the M.S. and Ph.D. degrees from Virginia Tech. In 1988, he joined Image Processing Technologies Inc. where he now holds the position of VP of Engineering. Since 1996 he has been teaching as adjunct professor at the Electrical and Computer Engineering Department of Virginia Tech. His interests include real time image processing and pattern recognition with applications in document capture and archiving.

About the Author—MORTON NADLER received the B.S. from CCNY and the M.S.E.E. from Illinois Institute of Technology for a pioneering thesis on time-domain synthesis of linear electric networks. His doctorate equivalent was earned at the Czechoslovak Academy of Sciences for work on high-speed arithmetic logic architectures in 1955–58. Among his innovations were carry look-ahead and SRT division and square root arithmetic logic.

He visited the Indian Statistical Institute in Calcutta (1959–1960) and then joined Cie des Machines Bull. Over a period of 11 years he investigated OCR covering innovations in pre-processors and hybrid recognition. He continued this line of investigation as an independent consultant for companies and institutions in Europe and the United States. In 1984 he joined the E.E. Dept. (now Electrical and Computer Engineering Dept.) at Virginia Tech. He founded Topologic Systems Corp. (now Image processing Technologies Inc.) in Blacksburg and Vienna VA to develop and market document capture and archiving systems.

View full text